Skip to main content
Skyvern’s MCP server and Skills let AI assistants (Claude Code, Cursor, Windsurf, Codex) control browsers, extract data, and run workflows through natural language. This page gets you set up.

Choose your path

Cloud (30 seconds)

Already have a Skyvern Cloud account? Jump to the MCP config for your client below.

Local (2 minutes)

Want to self-host? Run skyvern quickstart and it handles everything. See local setup.

Cloud setup

Get your API key from Settings in the Skyvern dashboard, then pick your client.
claude mcp add-json skyvern '{"type":"http","url":"https://api.skyvern.com/mcp/","headers":{"x-api-key":"YOUR_SKYVERN_API_KEY"}}' --scope user
That’s it. Ask your assistant to open a browser and navigate to any site.
Verify it works — paste this into your AI assistant: “Use Skyvern to create a browser session, go to news.ycombinator.com, and extract the title of the top post.”

Local setup

skyvern quickstart is the one-command path for self-hosting. Run it from your project or repo root:
pip install skyvern
skyvern quickstart
The interactive wizard will:
  • Set up a PostgreSQL database (detects local install or uses Docker)
  • Configure your LLM provider (OpenAI, Anthropic, Gemini, Ollama, etc.)
  • Write MCP config for your AI client (Claude Code, Cursor, Windsurf)
  • Install bundled Skills into .claude/skills/ (Claude Code only)
If you only want to configure without starting services, use skyvern init instead, then start the server manually:
skyvern run server
For the full local setup guide including browser modes and environment configuration, see Run with a local browser.

What gets installed

MCP Server

33 tools across 6 categories that your AI assistant calls automatically:
CategoryWhat it does
Browser sessionsOpen, manage, and reuse browser instances
Browser actionsNavigate, click, type, scroll — via natural language or selectors
Data extractionPull structured data from pages, take screenshots, run JavaScript
ValidationAI-powered checks on page state (returns true/false)
CredentialsLook up stored credentials for login flows
WorkflowsBuild and execute multi-step automations
Full reference: MCP Server

Skills

Markdown files that teach your AI assistant Skyvern patterns. Skills are not MCP tools — they’re documentation your agent reads for context on how to use the CLI and API effectively.
skyvern skill list            # List available skills
skyvern skill show skyvern    # Render a skill in the terminal
Full reference: CLI & Skills

Try it

Paste any of these into your AI assistant:
PromptWhat happens
”Use Skyvern to go to news.ycombinator.com and extract the top 5 post titles as JSON”Creates a session, navigates, extracts structured data, closes
”Create a Skyvern browser session and take a screenshot of example.com”Session creation + screenshot capture
”Run a Skyvern workflow that navigates to [url] and fills out the contact form with name ‘Jane Doe’ and email ‘jane@example.com‘“Multi-step workflow with form filling
”List my Skyvern credentials and use one to log into [site]“Credential lookup + authenticated login

Tips for better results

  • Tell your agent to use Skyvern MCP explicitly when you want browser automation
  • Be specific: “extract the price from the product page” beats “get info from the site”
  • For multi-step flows, ask your agent to use workflows or browser sessions with persistent state
  • Don’t hardcode passwords — use skyvern credential to store them securely
  • Use --json output when composing CLI commands in scripts

Next steps

Full MCP Reference

All 33 tools, config options, local mode, and troubleshooting

CLI & Skills

Browser commands, workflow management, and skill system

API Quickstart

Use the Python or TypeScript SDK directly